https://www.jstacs.de/index.php?action=history&feed=atom&title=Train_classifiers_using_GenDisMix_%28a_hybrid_learning_principle%29 2026-04-04T14:01:46Z Revision history for this page on the wiki MediaWiki 1.38.2 https://www.jstacs.de/index.php?title=Train_classifiers_using_GenDisMix_(a_hybrid_learning_principle)&diff=245&oldid=prev 2009-12-02T16:33:09Z

<p>New page: &lt;source lang=&quot;java5&quot;&gt; //read FastA-files Sample[] data = { new DNASample( args[0] ), new DNASample( args[1] ) }; AlphabetContainer container = data[0].getAlphabetContaine...</p> <p><b>New page</b></p><div>&lt;source lang=&quot;java5&quot;&gt;<br /> //read FastA-files<br /> Sample[] data = {<br /> new DNASample( args[0] ),<br /> new DNASample( args[1] )<br /> };<br /> AlphabetContainer container = data[0].getAlphabetContainer();<br /> int length = data[0].getElementLength();<br /> <br /> //equivalent sample size =^= ESS<br /> double essFg = 4, essBg = 4;<br /> //create ScoringFunction, here PWM<br /> NormalizableScoringFunction pwmFg = new BayesianNetworkScoringFunction( container, length, essFg, true, new InhomogeneousMarkov(0) );<br /> NormalizableScoringFunction pwmBg = new BayesianNetworkScoringFunction( container, length, essBg, true, new InhomogeneousMarkov(0) );<br /> <br /> //create parameters of the classifier<br /> GenDisMixClassifierParameterSet cps = new GenDisMixClassifierParameterSet(<br /> container,//the used alphabets<br /> length,//sequence length that can be modeled/classified<br /> Optimizer.QUASI_NEWTON_BFGS, 1E-9, 1E-11, 1,//optimization parameter<br /> false,//use free parameters or all<br /> KindOfParameter.PLUGIN,//how to start the numerical optimization<br /> true,//use a normalized objective function<br /> AbstractMultiThreadedOptimizableFunction.getNumberOfAvailableProcessors()//number of compute threads <br /> );<br /> <br /> //create classifiers<br /> LearningPrinciple[] lp = LearningPrinciple.values();<br /> GenDisMixClassifier[] cl = new GenDisMixClassifier[lp.length+1];<br /> //elementary learning principles<br /> int i = 0;<br /> for( ; i &lt; cl.length-1; i++ ){<br /> System.out.println( &quot;classifier &quot; + i + &quot; uses &quot; + lp[i] );<br /> cl[i] = new GenDisMixClassifier( cps, new CompositeLogPrior(), lp[i], pwmFg, pwmBg );<br /> }<br /> <br /> //use some weighted version of log conditional likelihood, log likelihood, and log prior<br /> double[] beta = {0.3,0.3,0.4};<br /> System.out.println( &quot;classifier &quot; + i + &quot; uses the weights &quot; + Arrays.toString( beta ) );<br /> cl[i] = new GenDisMixClassifier( cps, new CompositeLogPrior(), beta, pwmFg, pwmBg );<br /> <br /> //do what ever you like<br /> <br /> //e.g., train<br /> for( i = 0; i &lt; cl.length; i++ ){<br /> cl[i].train( data );<br /> }<br /> <br /> //e.g., evaluate (normally done on a test data set)<br /> MeasureParameters mp = new MeasureParameters( false, 0.95, 0.999, 0.999 );<br /> for( i = 0; i &lt; cl.length; i++ ){<br /> System.out.println( cl[i].evaluate( mp, true, data ) );<br /> }<br /> &lt;/source&gt;</div>

Keilwagen